AI’s Role in Transforming Radiology, with Dr. Kent Hutson of Radiology Partners

Episode Overview

Episode Topic: In this captivating episode of “Skeleton Crew – The Rad Tech Show,” we venture into the world of radiology and discover how Radiology Partners, a pioneering force in the field, is harnessing the power of artificial intelligence (AI) to transforming radiology and enhance patient care. Join us as we explore the fascinating conversation with Dr. Kent Hutson, Director for Innovation Clinical Operations at Radiology Partners, to unveil the remarkable impact of AI on radiology.

Lessons You’ll Learn: In this enlightening episode, you’ll uncover the vital role of AI in radiology and how it’s revolutionizing the way radiologists interpret medical images and produce reports. Dr. Hutson highlights the importance of AI in enhancing diagnostic accuracy, reducing radiologist burnout, and even identifying critical conditions faster, ultimately leading to improved patient outcomes. You’ll also gain insights into the challenges and opportunities of integrating AI into medical practices.

About Our Guests: Dr. Kent Hutson is a neuroradiologist and the Director of Innovation Clinical Operations at Radiology Partners. With a background in both computer science and medicine, he offers a unique perspective on the intersection of technology and healthcare, shedding light on the transformation AI is bringing to transforming radiology.

Topics Covered: In this engaging conversation, Dr. Hutson shares his journey and expertise, from his roots in computer science to becoming a radiologist. He explains the innovative AI solutions Radiology Partners has developed to optimize radiology workflows, improve patient care, and reduce radiologist burnout. The discussion touches on the AI applications for detecting critical conditions, the need for constant model monitoring, and the potential for AI in preventative care, providing a comprehensive view of the evolving landscape of transforming radiology in the age of artificial intelligence.

Our Guest: Exploring Radiology Advancements with Dr. R. Kent Hutson, MD, CPE

Follow Radiology Partners on

• Twitter
• LinkedIn
• YouTube
• Instagram
• Facebook

Episode Transcript

Dr. Kent Hutson: When COVID hit, we had models for chest x-ray for pneumonia, but nothing that would identify that. Very quickly People adapted and built models to recognize COVID pneumonia on chest x-ray. When you buy, when you start using AI, it’s not just a, oh, I’ll take this and start using it and it’ll work forever. No, it’s a process. You have to monitor it and you have to make sure that it’s still performing at the same level as when you started. Because of that, it’s very important to have a partner that can do that monitoring for you and make sure that it’s still functioning.

Jennifer Callahan: Welcome to the Skeleton Crew. I’m your host, Jen Callahan, a technologist with ten-plus years of experience. In each episode, we will explore the fast-paced, ever-changing stuff. That’s completely crazy field of radiology. We will speak to technologists from all different modalities about their careers and education. The educators and leaders who are shaping the field today and the business executives whose innovations are paving the future of radiology. This episode is brought to you by xraytech.org. If you’re considering a career in x-ray, visit xraytech.org. To explore schools and to get honest information on career paths, salaries and degree options. Hey, everybody.

Welcome back to the Skeleton Crew. My name is Jen Callahan, and I’m here today with an innovator in the field of radiology. I have Dr. Kent Hutson with me. He is the director for Innovation Clinical Operations at Radiology Partners. Kent, thanks so much for being with me today.

Dr. Kent Hutson: Thanks for having me Jen.

Jennifer Callahan: Could you just give me maybe a brief background of yourself, how you got yourself to where you are today?

Dr. Kent Hutson: Sure, so I’m a neuroradiologist, but before medical school, my original degree was computer science. So then I went to med school, then did a medical informatics fellowship, and then I went to radiology residency. I’ve had my hands in both computer science and radiology for quite a while now. About five years ago, joined Radiology Partners and at that time was involved in information technology, which is more on the clinical side. But I’ve always been interested in and have always done things in artificial intelligence. My postdoc fellowship was an artificial intelligence project back almost 30 years ago when started making an artificial intelligence department and adding the capabilities to do that. Very excited about the opportunity to join that team and so I’ve been with that team for about two and a half years now.

Jennifer Callahan: So before we delve into the AI portion of Rad Partners, could you just give maybe an oversight of what type of company Rad Partners is that? You have this innovation going on, but you’re also too into the clinical field, correct?

Dr. Kent Hutson: Sure. First off, if you use the word company to describe red partners, we have to put a dollar in the jar. We’re not a company. We’re a practice. We’re the largest practice in the US, about 3600 radiologists and we’re read for all 50 states. This year. We’ll read a little over 50 million exams. The way we’re set up is we have local practices that are in different physical locations and they function like you’d expect a radiology practice to function. The difference is that all the support functions like HR and IT and everything else that goes along with running a radiology practice is all done at the national level. So that’s done at the RP level. What I do, I’m a Teleradiologist, so I work for the Matrix Practice that does Teleradiology and reads a lot of our local practices. But then at the same time, I also spend part of my time doing the AI stuff at the national level.

Jennifer Callahan: So the local practices that you have, do they have their own offices, or they’re incorporated into into health systems?

Dr. Kent Hutson: Yeah, they’re governed locally, so they each have a local practice board that makes decisions about scheduling and the vacations and compensation and all that kind of stuff, just like a group would if they weren’t part of a larger practice. The difference is that all that other overhead stuff is managed at that national level, so we can do it much more efficiently with the economies of scale.

Jennifer Callahan: So let’s get into your expertise then. The AI portion, right? What are you currently have going on that is fun and exciting for yourself that you might want to do a show and tell on?

Dr. Kent Hutson: Well, RPED has a number of different AI deployments and for different uses. When you think about radiology AI, the first thing that comes to mind is the computer is reading my CTA scan or something like that. But actually the the first thing that we deployed was natural language processor that looks at our reports while we’re dictating it, it analyzes it for things like billing compliance, which is kind of boring, but better than that, it looks for things that need some kind of best practice recommendation. So for example, I’m reading a CT, the chest, I dictate that there’s a seven-millimeter nodule in the right upper lobe. And this AI system that’s running real-time while I’m dictating puts up an alert saying there’s a recommendation that you should get follow up in X number of months or get a biopsy or whatever the recommendation is. And those recommendations are based on looking going through the literature, the society recommendations for whatever that particular subspecialty is that have been reviewed by a group of our radiologists to say, okay, we’re going to adopt these standards, and then instead of all of us having to memorize all of those standards and all those little details from those tables, the AI presents us with the answer. It says, oh, okay, you need to recommend follow-up in X number of months. I see that on my report. I click the thumbs up button and it inserts it into my report.

Jennifer Callahan: So you have the choice to include it or not include it?

Dr. Kent Hutson: Right and with any kind of AI that we’re using in clinical medicine, the physicians should always have the final say as to what happens. We give the radiologists that choice. Do you use this or not? Because there might be some other history that it’s not in that report that you know about that I know about, that it would change that recommendation. Our compliance with best practice recommendations went from less than 10% success rate up to 80 plus percent rate almost overnight, so huge difference.

Jennifer Callahan: Yeah, something like that. I can only imagine. Extremely helpful. Your mind’s going in so many different directions, looking through the images. How many other protocols do I need to remember that I need to include this at the end of the dictation? So that’s amazing and probably extremely helpful for the radiologists that are utilizing it.

Dr. Kent Hutson: Yeah, and it gives us a much higher quality report in the end so that overall our whole practice delivers higher quality reports because of the use of AI. Another example, something we’re doing, we’ve deployed AI that is also watching the reports. So it has nothing to do with the images, watching the report. And when we dictate the findings, when we get down to the impression it automatically generates what it thinks should be the impression based on my previous reports. So thousands of my previous reports have been run through the model so that the system learns how I say things, how I format things, what I like to put in my impressions.

And then based on the findings in that report and training from thousands and thousands of other reports that aren’t mine, it generates an impression. Now, this works really well with a lot of the teleradiology stuff I do because I’m mostly doing emergency room stuff. So those are usually pretty straightforward cases and it’s not maybe changing the quality of what we produce, but we talk about improving burnout. It reduces burnout because it reduces the steps you have to take the workload on you and it just makes you feel better about what you’re doing at the end of your shift. And it saves time too, because the computer can generate that impression faster than what I could dictate all the impression, right?

Jennifer Callahan: And again, you have final say of what the impression will be. So say if you read it and there’s something that you might want to slightly tweak, you have the ability to do that.

Dr. Kent Hutson: Right and then just make the edits if I need to and then I sign. All of our reports are done through natural language processing and voice recognition. So we dictate a report, we sign it right then and it goes to the referring physician right away.

Jennifer Callahan: All right. So Radiology Partners is out there kind of known as transforming radiology or on a mission to transform that. What exactly does that mean for the rest of us who are learning about you?

Dr. Kent Hutson: So there are many ways that we’re trying to transform radiology, but I’m going to touch on the stuff because that’s what I do. We’re becoming an AI-enabled radiology practice. So what does that mean? The next part of our deployment we haven’t gotten to is the image part. So, for example, we built a cloud-native cloud-based orchestration platform that takes the images for the CT scan, whatever it might be, and looks at those images and says, okay, this is a CT of the head. I need to send it to this vendor, this model, this AI model, run it on that and get the result back and get the result back to the radiologist. A good example of this is you come into one of our emergency rooms where we read, right? And if you get a CT scan of the head for a car wreck, something like that, the study gets sent off to an AI model process to identify if there’s any kind of intracranial hemorrhage. So if there’s a bleed on that scan that throws a flag, and then that study gets moved up to the top of our list. So I’m sitting here cranking through studies and all of a sudden AI picks up a possible bleed, moves it to the top of the list.

And that’s the next study that I read. If there’s something wrong, there’s something critical wrong. That study is going to get read before others because normally it would just sit in the queue and it might be, you know, 20 or 30 minutes before it gets read, but not in this case. Same kind of thing with pulmonary embolism and a couple of other really high-value critical findings. And that’s just the beginning. And we find that using that kind of a system, we’re not only improving the turnaround time for those critical patients but also improving the detection rate. So we find that radiologists working with AI both together produce higher quality reports, higher quality results than than either alone. In some cases, it makes a significant difference. This year I mentioned that we’re going to read a little over 50 million exams, right? Our AI orchestrator is going to process a little over 20 million exams this year and we’re on target to beef that up and hoping that next year we’ll probably have over 40 million exams that we’re processing through AI. So we’re applying artificial intelligence live clinically right now to millions of exams.

Jennifer Callahan: I assume that if you’re reading that many exams, that the amount of radiologists that you have within the practice has to be a good number, right?

Dr. Kent Hutson: We have about 3600 radiologists.

Jennifer Callahan: And that’s spanning 50 states, correct?

Dr. Kent Hutson: Right.

Jennifer Callahan: Would you say a good majority of them are on-site or are remote or it’s kind of just a mix of both?

Dr. Kent Hutson: Well, Covid kind of changed that, didn’t it? Because it used to be that if you were in the town, then you were on site, right? A radiologist like me, I’m always off-site, obviously. But we’re finding now that a lot of our local practices have opted to. Do kind of a mixed thing where some days they’re on-site, you have to be there to do procedures, right? So, you know, if you’re doing procedures, you’re on site. But then at other times, they’re working from home just like the rest of the workforce in the country. Covid kind of changed things so that now we’re a mixed bag of working on-site, working from home.

Jennifer Callahan: So AI is definitely we can tell by what you’ve just said, a game changer in terms of radiology and workflow. Where do you see it kind of leading in the future for your practice and for the rest of the radiology world?

Dr. Kent Hutson: We’re really in a transition phase right now. 30 years ago, the transition was from Film to PACS and that was a game changer. Then we went live at my hospital. Back then, one of my partners came up to me and said, You’ve extended my professional life at least ten years. This is great. And we know now that over the past three decades, PACs has changed radiology tremendously. Just the ability to work from home, for example, the ability to cover multiple sites overnight so that the gets a report, a final report within 20 or 30 minutes in the middle of the night before PACS was unheard of. We’re in that kind of transition right now with AI right now, the PACs, viewers that we use to look at the studies, and all that, they’re not very well integrated with the AI. We’re just beginning to see that change. There are a couple of PACs, vendors that are making that change, but it’s going to take a little bit of a little bit of time. The standards around how to distribute results, that is in its infancy. It’s there, but it hasn’t been adopted widely. Same kind of thing happened three decades ago with Dicom. Dicom was born and that’s what facilitated the creation of PACs.

Dr. Kent Hutson: That’s the way we could do it because now all of a sudden you could send images around all over the place. We’re just beginning to get to that point with AI and radiology, where the results are standardized, can be standardized, and can be sent around and incorporated into the viewers so that you don’t have kind of some kind of separate widget off to one side that the radiologist has to look at in the future. The AI will be integrated into the viewer and things that right now take a lot of time will be completely different. An example if I see a nodule, I have to measure it. What if I could just say, Hey, there’s a nodule on the right upper lobe and AI has already identified it, has already measured it has already characterized it using a database of malignant versus benign. And so it can characterize that for me and insert that information into my report. I don’t have to do anything else. I don’t have to do all that measurement and spend all that time doing that. In the future, we’ll see more integrated into the workflow into what we do. So it becomes just a part of what we do.

Jennifer Callahan: So your AI, do you have it kind of just infrastruct it into your current practice or is it something that’s out there for other companies to be using or say, other radiology practices or health systems to be using?

Dr. Kent Hutson: We built our own infrastructure, I should say. We have some models that are our own a lot of models that we purchase from companies that develop the models, get FDA clearance, go through that whole process. But the infrastructure we built, we do sell that infrastructure. We do sell that as a service like AI, as a service to any health care systems that want it. So that’s there. And what we’re seeing in the in the market is the development of these marketplaces or platforms where companies are building the infrastructure and then they’re going out to companies just like we did, and deploying those models on that platform and then selling that platform as a service to the health care systems or radiology practices, whatever. And it’s kind of like you think about the App Store where you can if you want to use some AI, then you say, hey, I want this lung nodule thing. And they’ve already deployed it in their platform and then you pay whatever it is and use that AI. So as the market matures, we’ll see these marketplaces make it easier for healthcare systems to get into radiology, take advantage of it.

Jennifer Callahan: So do you see the role of AI being used in preventative care in the future or currently?

Dr. Kent Hutson: Yeah, actually, there’s some really good use cases where you can take what we call opportunistic screening. So for example, somebody gets a CT of the chest. Well, one of the things that you can do with that is use AI to determine how much calcification there is in the coronary arteries. So patient may have come in for a car wreck that CAT scan. And the AI says, oh, you have a high calcium score and then we can refer that patient to get appropriate treatment. Other things like identifying the bone mineral density on a study that really wasn’t originally designed for it, like a CAT scan or something like that. So identify, oh, this patient may well have osteoporosis. They need to go get screening, get proper treatment so that they reduce their risk for fracture, things like that.

Jennifer Callahan: Looking for the future for RP And AI. What’s the word that you guys are headed down?

Dr. Kent Hutson: We’re going to continue to integrate AI into what we do and make it part of our workflow, make it part of the practice. What we found is that once we deploy AI tools to a practice, to a group of radiologists, it doesn’t take very long for them to really get used to it and really rely on it. And they you can’t take it away. It becomes part of how they practice, part of what they do. Had one of our radiologists, when we did one of our very first pilots a couple of years ago, the comment was, Yeah, I kind of feel like it’s a little radiology resident sitting on my shoulder looking out for me. There’s so many benefits to it that I can’t see a world where we would go back and not use AI and not continue to incorporate AI into what we do.

Jennifer Callahan: How long would you say that it’s been incorporated into your practice?

Dr. Kent Hutson: We’ve been actively using it for we’re at five years now. That’s the one that watches the report for recommendations and that sort of thing. The imaging part we piloted about three years ago, we started doing that and we’ve been over the past few years when we’ve really taken off and started deploying that broadly. The deployment of AI should not be undervalued. Being able to deploy to a practice and getting the AI in front of the radiologists is a massive undertaking, not just from the technical side, but remember it’s a new tool. So they have to learn how the model works and not from a mathematical standpoint.

But okay, I know that this model is probably going to overcall fractures of the scaphoid or this model is going to have trouble. If there’s some motion artifact on this CAT scan of the head and it might call a hemorrhage on this because of the motion. So the radiologist has to understand what the artifacts are, what the pitfalls are, just like any other kind of imaging modality. Ultrasound can’t show you this. And you know that an MRI may have trouble with this because of like metallic artifacts, something like that. It’s the same kind of thing. The radiologist has to learn where the artifacts are, how the model performs in certain situations, and be able to account for that, because the models are not right all the time, that’s for sure. And ultimately, it’s the radiologist that has to make the decision about what’s going on on this particular exam. The AI is there to help, but it’s certainly not able to make a final diagnosis at this point.

Jennifer Callahan: So what advice do you have for radiologists that are out there that are integrating in with the AI?

Dr. Kent Hutson: The first thing I would say is it’s not the same thing as just buying a packs viewer or buying a 3D renderer or something like that, because every model that gets deployed as soon as it’s built, it’s going to start degrading. It’s not going to work forever. It’s trained on a certain data set, and sometimes we don’t even know what data set it’s trained on. As you go forward, you’re going to see there are going to be new disease processes and more commonly, new machines. So new CAT scanners, new MRI scanners have new acquisition techniques. And that model probably would not have been trained on those new images, on those new scanners. A great example is when COVID hit, we had models for chest x-ray for pneumonia, but nothing that would identify that very quickly. People adapted and built models to recognize COVID pneumonia on chest x-ray. So when you buy when you start using AI, it’s not just a, oh, I’ll take this and start using it and it’ll work forever. No, it’s a process. You have to monitor it and you have to make sure that it’s still performing at the same level as when you started. Because of that, it’s very important to have a partner that can do that monitoring for you and make sure that it’s still functioning. So we’re doing that internally, but that’s because we have the resources to build it internally and to do that monitoring ourselves. But most practices, most health care systems are not going to have that in-house expertise. So it’s good for them to look at these platforms, these marketplaces, and talk to them first and find a good partner. That’s a good match for them so that they have somebody that they can call on to do that model monitoring and make sure that the quality improvement process is done properly.

Jennifer Callahan: So say if I was Joe Schmo on the corner and you know, I am so interested in AI work at a radiology practice, I want to incorporate this into my practice, and or maybe I might be someone who is interested in getting into just working into AI. What’s your advice there? What would be the pathway to this?

Dr. Kent Hutson: So we’ve seen a number of different pathways from a radiologist or ad tech or maybe somebody who’s in it, but not in health care that have come into this field because it is pretty early. It’s still developing. My suggestion would be to start at SIIM, Society for Imaging, Informatics, and Medicine. So siim.org , SIIM has great courses to teach you about to kind of fill in the gaps. So if you’re an IT person with no health care experience, they’ve got courses for that. If you’re a healthcare person that needs the IT education courses, they’ve got that and they have a program that you can get certified for imaging informatics and they have a lot of content for AI and it gets you up to speed for that. Matter of fact, I just got back from many conference, which is a is put on by SIIM that is devoted entirely to artificial intelligence in medical imaging. And so for two days, that’s all we did. If you go to the regular SIIM conference, which is usually in the summer, they have a lot of good lectures on AI and IT and how to deploy it, how to purchase it, and all the things that we’ve kind of touched on during this discussion.

Jennifer Callahan: Well, That was an enlightening conversation. Kent, thank you so much. Your Radiology Partners practice. I don’t want to have to put a dollar in the jar. Seems like it’s doing some great stuff out there, and especially with the AI being incorporated into everything, definitely aiding your radiologists in their workflow and decreasing burnout, and making sure being preventative as well. So thank you for taking the time with me today. Everybody, this is Kent Hudson from Radiology Partners coming from where are you at California, right?

Dr. Kent Hutson: Palmer Lake, Colorado.

Jennifer Callahan: Oh, Colorado, my goodness. Excuse me. So Colorado. Thank you again, Kent. I really appreciate it. And everybody will catch you on the next episode here. Thanks for being with us again. We’ll see you later.

You’ve been listening to the Skeleton Crew brought to you by xraytech.org. Join us on the next episode to explore the present and the future of the rad tech career and the field of radiology.